SaltGAN: A feature-infused and loss-controlled generative adversarial network with preserved checkpoints for evolving histopathology images

Olaide Nathaniel Oyelade, Hui Wang, Sunday A. Adewuyi

Abstract: Highlights•Architectural design of adversarial neural networks comprising of a generator and discriminator.•A convolutional neural network (CNN) for detection and extraction discriminant features and for classification purposes was designed.•Application of a layer-wise skip connection of convolutional operations from the CNN architecture to corresponding layers in generator network.•Formulation of an improved loss function for both the generator and discriminator neural adversarial networks to improve the min–max search.•Design of ranking system using multi metric approach (MMA) for evaluating and restoring checkpoints to preserve learning curve of the adversarial model.